<h1 id="welcome!-2ptrfk7oq">Welcome!</h1>This is an experiment in OPEN finetuning. This means that I will give you everything I used to create this finetune.Warning: do not use default CFG of 7! cfg=3 to 6 is typically bestThe training dataset is a subset of <a target="_blank" rel="ugc" href="https://huggingface.co/datasets/ppbrown/danbooru-cleaned">https://huggingface.co/datasets/ppbrown/danbooru-cleaned</a> More details on that down below<h2 id="backstory-hvrc22t77">Backstory</h2>From an idealist standpoint, I wanted to create a finetune that was maximally compatible with everything. So I always tried to train only on SDXL base model.However, doing this round of finetuning experiments, I was getting HORRIBLE output from base sdxl. The colors were fine, but for some reason, the human likenesses were waayyy off.So I abandoned my earlier principles of compatibility for now, and went for a model that does humans better: Juggernaut.Sad story: turns out, the latest version of juggernaut actually does anime pretty well :-(But the huggingface "Juggernaut XL" model i was playing with did not, so from where I started, it is still technically an improvement. (it is juggernaut 9)<h2 id="journal-19y1u84an">Journal</h2>I log some of the details of my journey, at<a target="_blank" rel="ugc" href="https://civarchive.com/articles/7521/my-saga-of-fine-tuning-sdxl-models">https://civarchive.com/articles/7521/my-saga-of-fine-tuning-sdxl-models</a><h2 id="settings-xavoraq9d">Settings</h2>I used OneTrainer on an RTX 4090, with the following high level settings:input type: bf16training type : bf16LR: 1e-05, constant. Batchsize=1, 100 epochsEMA step: 100, EMA on CPU, XFORMERS attention, Gradient checkpointing on.NO text model trainingI set resolution to 1024 in training settings, but did NOT enable resolution override in the "concept" definitionThese settings gave me a speed of around 2IT/s and used around 16GB of VRAM<h2 id="full-dataset-details-bx8e5s4q3">Full dataset details</h2>You can use the scripts in<a target="_blank" rel="ugc" href="https://huggingface.co/datasets/ppbrown/danbooru-cleaned/tree/main/SUBSETS">https://huggingface.co/datasets/ppbrown/danbooru-cleaned/tree/main/SUBSETS</a>using the input file at<a target="_blank" rel="ugc" href="https://huggingface.co/datasets/ppbrown/danbooru-cleaned/blob/main/SUBSETS/files.onegirlscene605.txt">https://huggingface.co/datasets/ppbrown/danbooru-cleaned/blob/main/SUBSETS/files.onegirlscene605.txt</a> This will download the relevant .jpg files, AND the required .txt tagging files<h1 id="license-y1tl9ypb1">License</h1>Reminder: this model is Juggernaut based, so it is subject to the juggernaut license!!

Files with SHA256: bf466b51cabce9d1bbde6d5f706247596c881b87522e3b13df6ebe300ab00bce